Happy Sunday Fam! Two of the three biggest AI labs shipped models this week that you can't actually use. Google killed the Chromebook brand on a Tuesday. The Musk-vs-Altman jury starts deliberating tomorrow. And on the AI-only social network Meta acquired earlier this year, the chatbots have apparently started posting therapy threads. Welcome to the week.

Two stories ran in parallel this week. They look unrelated. They aren't.

Monday: Anthropic announced Mythos, a model it calls "strikingly capable" at hacking and offensive cybersecurity — too capable, the company says, to release publicly. Instead, Anthropic is spinning up Project Glasswing, a coalition with Amazon, Apple, Google, and Microsoft to figure out responsible deployment. The same day, Google's threat-intel team warned that attackers are already using AI to write zero-day exploits in the wild. Glasswing, in other words, is not a theoretical exercise.

Also Monday: OpenAI confirmed it will give the European Union limited access to GPT-5.5-Cyber, a security-focused variant of its frontier model — but only to vetted teams, and Anthropic is still holding Mythos back from the EU entirely.

Here's the shift: for the first time, the labs are publicly disagreeing about how to handle a class of capability they all agree is dangerous. Anthropic is choosing coalition gating. OpenAI is choosing sovereign disclosure. Neither is just shipping the model. This is governance happening in real time, without legislation, run by four or five companies whose valuations exceed most national budgets — Anthropic alone is reportedly raising $30–50B at a $950B valuation this week.

If you build on these APIs, this matters: the availability of a model is now a strategic decision, not a technical one. The model you depend on may exist but be unreachable for your industry, region, or use case. We'll start tagging tools in the directory by underlying provider and gating status next month. The era of "just pick a model" is wrapping up.

Own a tool that you'd like to see on our site or newsletter? Please Submit a tool or reach us by replying to this email or at [email protected].

Own a tool that you would like to see on our site or newsletter please Submit a tool or reach us by replying to this email or at [email protected]

OpenAI's former CTO is having a week. Her new company, Thinking Machines, unveiled what it's calling "interaction models" — AI that continuously interprets audio, video, and text simultaneously and responds in real time, without waiting for a send button. Think less "chatbot" and more "colleague who never lets you finish your sentence." It's the most concrete pitch yet for a post-chat AI interface, and it lands the same week Google announced Gemini Intelligence will move autonomously across apps. The shift from "AI as a text box" to "AI as ambient agent" is happening in real time, and Murati just planted a very large flag in the middle of it.

Coinbase cut 14% of its workforce this week and explicitly blamed "AI productivity improvements." Meanwhile, 76% of companies now have a Chief AI Officer — and the first place most of them are pointing the new hire is finance and accounting. Our 2026 collection rounds up the tools actually doing the work: AI bookkeepers, close-process automation, forecasting copilots, AP/AR agents, and the modeling assistants finance teams are quietly standardizing on this year.

Explore the Collection

Build sites at the speed of AI — without losing design control

Webflow is the visual web platform marketing teams ship on when "vibe-coded prototype" isn't going to survive a brand review. You design in a real visual canvas, the platform writes the clean HTML/CSS/JS for you, and your CMS, SEO, hosting, and localization come standard — no separate stack to babysit.

In a week where every announcement was about AI agents acting across your apps, Webflow is the layer that makes the front-end side of that loop actually presentable. Build the page, plug in your CMS, hand the keys to marketing, and let the agents handle the rest.

Check out Webflow →

If you want to share your company or product with fellow AI enthusiasts before we're fully booked, submit your tool here and sign up.

If you want to share your company or product with fellow AI enthusiasts before we’re fully booked, submit your tool here and sign up!

55 dead AI tools in 2026 alone — the worst year on record. Our AI Graveyard is now sitting at 166 entries total: 96 shut down, 69 acquired and sunset, and 1 acquired but still shipping. A few notable headstones added this month:

Browse the full graveyard →

The University of Exeter dropped a paper this week confirming chatbots will agree with your worst theories because that's how they're trained. Run this against any AI-generated draft before you ship it.

Especially useful before publishing anything generated by GPT-5.5 — which posts the highest accuracy on AA-Omniscience benchmarks and an 86% hallucination rate. Most confident wrong answers you'll ever see.